Full bandgap defect state characterization of β-Ga₂O₃ grown by metal organic chemical vapor deposition

Beta-phase gallium oxide (β-Ga₂O₃) is a promising candidate material for applications in high-power and radio frequency (RF) electronics due to its wide bandgap of ∼4.5–4.8 eV,^1–3 the ability to achieve (Al_xGa_1−x)₂O₃/Ga₂O₃ heterojunctions,⁴ its ease of n-type doping,^5,6 and the availability of large area, melt-grown β-Ga₂O₃ substrates. Theoretical predictions suggest the possibility of achieving very large breakdown fields of ∼8 MV/cm^7,8 and figures of merit that can exceed those of GaN and SiC.^8,9 The availability of native β-Ga₂O₃ substrates enables homoepitaxial growth of β-Ga₂O₃ device layers, which implies high device reliability in future applications since high concentrations of dislocations in epitaxial devices are not anticipated. As a result of these properties, there has been a surge in research efforts focused on β-Ga₂O₃ over the past several years. With regard to epitaxial structures, β-Ga₂O₃ grown by molecular-beam epitaxy (MBE) is being widely explored, with efforts on growth optimization, doping, heterostructure development, device characterization, and defect investigations all ongoing.^10–12 MBE-based devices have yielded promising results, including δ-doped metal-semiconductor field-effect transistors (MESFETs) with cutoff frequencies of 27 GHz,¹³ fin-shaped field-effect transistor (FINFET) devices with breakdown voltages exceeding 1.6 kV,¹⁴ high-power field plated Schottky barrier diodes and rectifiers,^15–17 high 2DEG charge densities in (Al_xGa_1−x)₂O₃/Ga₂O₃ modulation-doped field-effect transistors (MODFETSs),¹⁸ and superior power switching figure of merits in enhancement mode β-Ga₂O₃ transistors.¹⁹ While MBE-grown gallium oxide materials and devices have continued to advance in performance at an accelerated pace for several years, β-Ga₂O₃ epitaxial layers grown by metal-organic chemical vapor deposition (MOCVD) are at a comparatively earlier stage of development.^20,21 Despite this, very promising early reports have already established that MOCVD-grown β-Ga₂O₃ can produce transport characteristics at materials levels that are at least on par, if not exceeding, state-of-the-art MBE-grown bulk electron mobility values^22,23 with room temperature electron mobilities of up to 184 cm²/V-s reported for lightly Si-doped epitaxial β-Ga₂O₃ layers.²⁰ This impressive result implies a low concentration of defects for these MOCVD films. However, unlike the case for β-Ga₂O₃ grown by both MBE and bulk-growth methods where defect states in the bandgap have now been extensively reported,¹⁰ only sparse information currently exists regarding deep levels in MOCVD-grown β-Ga₂O₃, and those reports only cover a limited portion of the bandgap.²⁴ Determining the entire deep level distribution in the bandgap is necessary to identify key defects that cause issues impacting device performance, such as carrier compensation, recombination-generation, trapping, scattering, and so forth. Comparison of the deep level defect distribution with reports for β-Ga₂O₃ grown by other methods,^10,25 and also comparing to theoretically calculated energy levels,^5,26,27 can give clues regarding their physical sources and, as such, can provide guidelines for continued materials optimization. This work reports the energy and concentration profiles of bandgap states within MOVCD-grown β-Ga₂O₃ using a combination of Deep Level Optical Spectroscopy (DLOS), Deep Level Transient (thermal) Spectroscopy (DLTS), and Admittance Spectroscopy (AS).

Samples for this study were grown in an Agnitron Agilis R&D MOCVD system using TEGa (triethylgallium) and O₂ precursors. Test layers were grown to a target thickness of 1 μm using a nominal Si target doping of 1 × 10¹⁷ cm⁻³, which was confirmed by secondary ion mass spectrometry (SIMS) measurements. Intentional Si doping was used to ensure a uniform, well-controlled, and low concentration doping profile to enhance trap spectroscopy analysis. The layers were grown on commercially available Sn-doped (010) EFG (edge-defined film fed grown) substrates from Novel Crystal Technology (Tamura) at a growth temperature of 880 °C using a growth rate of 0.7 μm/h. As noted above, MOCVD-grown lightly Si-doped β-Ga₂O₃ layers using these same growth conditions revealed a room temperature Hall electron mobility of 184 cm²/V-s (4984 cm²/V-s at 45 K), with an electron concentration of 2.5 × 10¹⁶ cm⁻³ at 300 K.²⁰ Complete details of the MOCVD growth can be found in the work of Feng et al.²⁰ Once grown, the structures were processed into Ni Schottky barrier diodes for subsequent electrical and defect spectroscopy measurements using standard photolithographic processes.^10,25 Prior to the Ni Schottky metal deposition, the substrate was cleaned using a standard solvent clean by acetone, isopropyl alcohol, and de-ionized (DI) water. Ni was, then, deposited by electron beam evaporation to a thickness of 8 nm, thin enough to allow light penetration for DLOS studies but also robust enough for DLTS and admittance spectroscopy measurements. The Schottky contact area was 8.41 × 10⁻⁴ cm². A mesa etch was performed using BCl₃/Ar chemistry to isolate the devices. Finally, an ohmic stack of Ti/Al/Ni/Au, which is reported to have a low specific contact resistivity,⁶ was deposited on the front side after a mesa isolation etch was performed. Full device processing details have been previously published, following our standard approach for DLOS and DLTS studies of β-Ga₂O₃ Schottky diodes.^10,25

Test structures were screened to ensure high quality devices were being used via the following methods: Hall effect, current-voltage (IV), capacitance-voltage (CV), and internal photoemission (IPE). Figure 1 shows representative IV, CV, and CV-extracted net ionized doping concentrations, all of which revealed consistent and high quality devices suitable for defect spectroscopy. Diode ideality factors at 300 K ranged from 1.02 to 1.07 for the 10 devices fabricated on this substrate, which is consistent with a nearly ideal thermionic emission-controlled Schottky diode. IPE measurements across all 10 diodes were very consistent, revealing a Schottky barrier height of 1.4 V ± 0.1 V. The extracted net ionized doping concentration from C-V was 1.2 × 10¹⁷ cm⁻³ close to the target value noted above. A separate sample grown for Hall studies under identical growth and doping conditions revealed this layer to have an electron mobility of 152 cm²/V-s at 300 K, which follows the expected trend with carrier concentration based on the earlier lightly Si-doped results of Feng et al.²⁰ 

With the quality of the test devices established, defect spectroscopy could commence. Following our prior work on MBE and EFG-grown β-Ga₂O₃, both DLTS and DLOS measurements were used to probe the full range of bandgap states. Complete details of both DLTS and DLOS measurements can be found elsewhere but are briefly outlined here.^28,29 DLTS measurements were performed using a fill pulse bias of 0 V with a 10 ms duration to fill trap states. To monitor the thermally stimulated carrier emission processes, a quiescent reverse bias of −2 V was used. The capacitance transients were recorded over a temperature range from 80 K to 400 K in steps of 0.1 K. The temperature-dependent capacitance transient spectra were analyzed using a conventional double boxcar method across a wide range of rate windows from 0.8 s⁻¹–2000 s⁻¹. With these measurement conditions, the thermally stimulated emission based DLTS method typically can provide trap information for states with activation energies of approximately up to 1 eV. The remainder of the bandgap was probed using DLOS, wherein optical stimulation of carriers from deep levels in the bandgap is used to overcome the carrier freeze-out limitation issue for DLTS for states that exist with activation energies greater than 1 eV, all the way to the bandgap energy. In our DLOS setup, photoemission transients were measured for 300 s as a function of incident photon energy using a spectrally resolved, monochromatic sub-bandgap light source, at 300 K. Two different light sources, an Ushio No. 1000486 Quartz Tungsten Halogen (QTH) lamp (600 W) and a Newport 6269 Xenon Arc lamp (1000 W), were dispersed through a high resolution monochromator to provide a tunable, high resolution light source ranging in energy from 0.5 eV to 5.0 eV in 0.02 eV increments. Trap filling and quiescent biases were the same as used for the DLTS measurements except the fill pulse duration was increased to 10 s, as discussed in prior publications.^10,25 The steady state photocapacitance (SSPC) as a function of incident photon energy was used to extract concentrations of DLOS-detected traps, with the SSPC onset energies being indicative of the trap energies. More precise DLOS trap energy and associated Frank–Condon energy levels were extracted by modeling of the photocapacitance transients through fitting to the Passler model of optical cross sections.³⁰ A more detailed description of the extraction of precise energies associated with DLOS-detected states has been published previously.¹⁰ 

DLTS measurements were performed on multiple devices to ensure consistency in the results, using a standard boxcar analysis of the temperature-dependent capacitance transients for ten rate windows ranging from 0.8 s⁻¹ to 2000 s⁻¹. A representative DLTS spectrum for the 4 s⁻¹ rate window is shown in Fig. 2(a), revealing the presence of a single trap. The Arrhenius plot for this trap was obtained using all of the rate windows and is shown in Fig. 2(b), from which the trap energy and capture cross section values were determined to be E_C-0.4 eV and 1.5 × 10⁻¹⁴ cm², respectively. The concentration of this trap was calculated to be 3 × 10¹³ cm⁻³, taking into account the so-called lambda effect, which accounts and corrects for non-uniform ionization of the E_C-0.4 eV trap throughout the entire depletion region at the bias conditions used.²⁸ From Fig. 2(b), this state appears distinct from our previous DLTS studies made on both Ge-doped plasma assisted molecular beam epitaxy (PAMBE)¹⁰ and unintentionally doped (UID) EFG-grown materials.²⁵ Also shown in Fig. 2(a) is a simulated DLTS peak response calculated for an ideal, isolated, non-interacting trap state having the same energy level and capture cross section values as the measured trap.²⁹ The excellent fit to this simple model implies that the source for this trap is likely to be a simple point defect.

With DLTS establishing the trap spectrum in the upper region of the bandgap, we now turn to DLOS for the remainder of the bandgap. Figure 3(a) shows the steady state photocapacitance (SSPC) spectrum revealing two onsets indicated by the arrows, with a third lower energy SSPC onset magnified in Fig. 3(b). Overall, there are three positive photocapacitance onsets indicating traps near the specified energies. While the SSPC onset energies indicate the incident optical energies at which the photoemission affects the photocapacitance, fitting of the optical cross section data derived from the photocapacitance transients enables more accurate determination of each trap energy level and their associated Frank–Condon energy (DFC).^30,32 Figure 3(c) shows the optical cross section data fitted using the Passler model, from which energy levels and DFC values are obtained.^10,30,32 From this fitting, the three DLOS-detected states were determined to have energy levels of E_C-1.2 eV, E_C-2.0 eV, and E_C-4.4 eV, with associated DFC values of 0.45 eV, 0.48 eV, and 0.06 eV, respectively. The concentrations of these trap states at E_C-1.2 eV, E_C-2.0 eV, and E_C-4.4 eV were 1.3 × 10¹³ cm⁻³, 2.3 × 10¹⁵ cm⁻³, and 2.1 × 10¹⁵ cm⁻³, respectively. These three states closely match DLOS-detected states previously reported for β-Ga₂O₃ grown by MBE¹⁰ and EFG,²⁵ suggestive of common physical sources. There have been several efforts to explore physical sources of these states and their relative impact on material properties for MBE and EFG materials, and these are briefly discussed to assist in source identification, and differentiation, for the MOCVD materials studied in this work.^10,25 Our prior work has shown that both E_C-1.2 eV and E_C-2.0 eV traps are sensitive to high-energy neutron irradiation, each with different defect introduction rates.³³ Moreover, it was found that these two states are the primary compensating deep levels causing carrier removal after neutron irradiation. The sensitivity to radiation fluence implies that intrinsic physical sources, such as vacancies, self-interstitials, or possible point defect complexes, involving native defects are most likely responsible for these states. In fact, recent studies have shown a strong correlation between the E_C-2.0 eV state and the presence of 2V_Ga-Ga_i complexes based on a combination of high resolution electron microscopy studies and density functional theory (DFT) calculations.^34,35 That the E_C-2.0 eV trap concentration obtained from DLOS for the MOCVD β-Ga₂O₃ material here is approximately 20× less than what has been observed for MBE and EFG materials (discussed below and shown in Fig. 4) implies a dependence on growth method. Such a dependence on differences between MOCVD, MBE, and EFG growth conditions would not be surprising if a native defect source is linked to this state.

The SSPC spectrum in Fig. 3(a) also shows the presence of a negative slope starting near 3.2 eV, which is perceptible in the optical cross section data in Fig. 3(c). This feature has been occasionally observed in earlier DLOS studies on PAMBE¹⁰ and EFG materials for this state.²⁵ While the source of this feature is unclear, it is reproducible and prominent for the MOCVD material and, thus, merits discussion. The negative slope could mean either there is an increase in negative space charge or it occurs likely due to reduction in optical cross section magnitude explained hereafter. In DLOS, it is possible that when the incident photon energy is greater than half the bandgap for an arbitrary semiconductor (greater than approximately 2.4 eV for β-Ga₂O₃), there can be competition between electron emission to the conduction band and electron capture from (i.e., hole emission to) the valence band for a given state.³⁶ If the latter process becomes significant, the observed SSPC magnitude would result from a competition between the two processes. However, if the hole in β-Ga₂O₃ is not mobile, as has been widely predicted, and instead forms a small polaron that acts as a self-trapped hole,³⁷ there can be no change in the space charge due to this process. This is because the positive charge on the immobile hole will balance the negative charge due to hole emission from the trap to the valence band.³⁸ If the hole has some mobility, possibly due to hopping, then such an explanation would be feasible. More work is needed to understand this further. An alternative explanation is that we are simply observing the normal behavior of the optical cross section associated with this defect, where the optical cross section reduces in magnitude for energies higher than the maximum resonance point for a given energy level.^32,36,39

Moving now to the state at E_C-4.4 eV, we first note that this level has been observed in all DLOS studies of β-Ga₂O₃ to date, regardless of the growth method, and its concentration has not appeared to vary significantly across a wide range of samples grown under different conditions, as a function of doping, or even after high-energy neutron and proton irradiations.^10,25,33 This apparent invariance for the E_C-4.4 eV state has led to the speculation that the source for this feature might be related to a fundamental property of gallium oxide itself, including the possible role of self- trapped holes, which has been very tentatively suggested previously.⁴⁰ However, such an association is inconsistent with the observation seen here on the MOCVD materials, where a very large reduction in the concentration of the E_C-4.4 eV state concentration is seen. A comparison of SSPC spectra at the same scale for β-Ga₂O₃ grown by MOCVD vs our prior work on MBE and EFG materials is provided in Fig. 4. All measurements were performed under identical conditions so meaningful comparisons are established. It is very clear that all of the DLOS-detected states are greatly diminished in their concentrations for the MOCVD-grown material. Since bandgap states in the range of detection for DLOS are very likely to be acceptor-like in this n-type material, such low concentrations are consistent with the low concentration of total compensating acceptors (∼9 × 10¹⁴ cm⁻³) extracted from the transport studies published previously on the high mobility lightly Si-doped MOCVD material.²⁰ The large overall reduction in total trap concentration by approximately 10× for the MOCVD material is significant, given the similarities observed in prior studies, and is consistent with the measured high 300 K electron mobility of 152 cm²/V-s for this lightly Si-doped sample. Furthermore, with regard to the E_C-4.4 eV state, its significant reduction in concentration here, coupled with the lack of any dependence on high-energy particle irradiation observed in earlier work, implies that an extrinsic source may be responsible. While more work is needed to discern the source of the E_C-4.4 eV state, especially given its relative dominance in the deep state concentration profiles reported to date, this tentative association with an extrinsic defect source is the first significant correlation of this state with growth conditions.

While the combination of DLTS and DLOS can provide full coverage of states within the β-Ga₂O₃ bandgap, the presence of increased ohmic contact resistance at very low temperatures for our devices (below ∼80 K here) limits the applicability of DLTS in that range, making detection of very shallow traps (closer to E_C) difficult, especially for those states which have high carrier emission rates. This is a concern because recent transport studies on the MOCVD material have implied the presence of a deep donor state at approximately E_C-0.12 eV.²⁰ In an attempt to circumvent this issue, we resorted to admittance spectroscopy (AS) measurements since AS enables the observation of traps having relatively fast emission rates but at a higher measurement temperature, thereby circumventing the contact resistance issue faced during low temperature DLTS measurements. Admittance spectroscopy was performed at 50 mV AC bias, and the DC bias was fixed to 0 V. The measurement was performed at various temperatures ranging from 140 K to 260 K, as depicted in Fig. 5(a). Following prior work on admittance spectroscopy,^41,42 the derivative of capacitance as a function of measurement frequency reveals a peak value if a trap is present, where the peak frequency ω_p (inflection point), corresponds to the trap emission rate. Figure 5(b) demonstrates the measurement of an inflection point at 200 K, and it can be similarly applied for all of the measured temperatures. The change in temperature for AS will have a similar impact as that of DLTS and the emission rate will follow an Arrhenius relation, as shown in Fig. 5(c). Here, from Fig. 5(c), we do see the presence of a trap that has an activation energy of E_C-0.12 eV and a capture cross section of 5 × 10⁻¹⁷ cm². The concentration of this trap using AS can be calculated from the change in capacitance in Fig. 5(a). At lower frequencies, the measured capacitance is composed of the depletion capacitance and is affected by the charge contribution from the trap state, whereas at higher frequencies, the traps cannot respond. Hence, the difference between the low and high-frequency capacitance provides the capacitance due to the trapping contribution alone (i.e., ΔC), through which the trap concentration was found to be 3.1 × 10¹⁵ cm⁻³. The concentration and activation energy of this state are in good agreement with the values extracted from transport measurements made on lightly Si-doped β-Ga₂O₃ grown by MOCVD reported earlier.²⁰ Note that the AS data are also included in Fig. 4 and, as seen, are dominant in the MOCVD material. This correlation between trap spectroscopy and transport analysis reveals consistency between very different measurements, and work must now be done to explore the physical source for this defect state given its relatively high concentration compared with the other states seen by DLTS and DLOS in MOCVD-grown β-Ga₂O₃.

With the E_C-0.12 eV state clearly revealed in the Si-doped MOCVD material by AS, we decided to apply AS to β-Ga₂O₃ Schottky diodes grown using the UID EFG-grown²⁵ and Ge-doped β-Ga₂O₃ PAMBE-grown material,¹⁰ which we previously characterized by DLTS and DLOS. For the EFG material, AS revealed the same state, which is consistent with the AS work reported by Neal et al.⁴¹ However, there was no evidence of this state in the Ge-doped PAMBE material from these measurements. As we have previously reported for the EFG sample, SIMS revealed a background Si concentration on the order of 10¹⁷ cm⁻³ for the UID EFG sample, whereas SIMS showed no measurable Si concentration in the Ge-doped PAMBE-grown material. Whether this state is related to the presence of Si, associated defects, or even site competition between the Ga_I and Ga_II sites of the β-Ga₂O₃ lattice is under study currently.

In summary, a comprehensive investigation of the bandgap states in MOCVD-grown β-Ga₂O₃ was completed using a combination of DLOS, DLTS, and admittance spectroscopy (AS). A large reduction in overall trap concentration was observed compared with all prior studies to date on the full bandgap spectrum of defects made on materials grown by PAMBE¹⁰ and EFG.²⁵ The dominant state for the MOCVD material is a relatively shallow state at E_C-0.12 eV, which was detected by AS. Its presence matches findings from previous transport studies made on the MOCVD material.²⁰ Unlike previous DLOS studies, the E_C-4.4 eV state is no longer the dominant deep state, implying that its source might be extrinsic in nature. Furthermore, DLTS revealed a previously unreported state at E_C-0.4 eV, which exhibits ideal trapping characteristics suggestive of a simple point defect source. In general, all states previously associated with an intrinsic source, including the increasingly studied E_C-2.0 eV trap, are diminished in concentration. The findings discussed here are consistent with the high electron mobilities and very low acceptor-like compensating state concentrations recently reported for MOCVD-grown β-Ga₂O₃ produced in the same reactor under identical growth conditions. These results strongly suggest that MOCVD-grown β-Ga₂O₃ has great potential to enable high performance ultra-wide bandgap electronic and optoelectronic devices.

The authors acknowledge the funding support from Air Force Office of Scientific Research No. FA9550-18-1-0479 (Ali Sayir, Program Manager). The authors also acknowledge the funding support from the Department of the Defense, Defense Threat Reduction Agency, Grant No. HDTRA11710034 (Jacob Calkins, Program Manager). This work was also sponsored by the NSF Graduate Research Fellowship Program under Grant No. DGE-1343012. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funding provider.